This study evaluates the seasonal predictability of the tropical Atlantic variability (TAV), which influences the regional and global climate conditions significantly. Two leading TAV patterns, the northern tropical Atlantic (NTA) and southern tropical Atlantic (STA) patterns, are associated with distinctive events of sea surface temperature (SST) anomalies that last for more than a season. Currently, the skill of the TAV seasonal prediction is not high, possibly because of the model systematic error, inadequate observation and/or the intrinsic limit of the TAV predictability. To isolate the last factor, a series of ensemble predictability experiments in a "perfect" model setting have been conducted using the global coupled community earth system model (CESM), Version 1.1. The analysis of an 86-year control simulation shows that CESM reproduces the annual cycle in the tropical Atlantic and the NTA and STA patterns realistically. A set of major NTA and STA events is selected from the control simulation and ensemble predictability experiments are conducted to predict each of them from monthly to seasonal lead times with small perturbations of the atmospheric initial states. The correlation and root mean square error (RMSE) of the ensemble mean prediction, as well as the intra-ensemble spread, are used to assess the prediction skill and reliability quantitatively. It is found that the model can predict these NTA events skillfully up to 9 months and STA up to 4 months. Composite analysis is conducted to identify the sources of predictability, including the air-sea interactions in the tropical Atlantic and the remote forcing from the El Nino-Southern Oscillation (ENSO). The surface latent heat flux anomalies, generated by surface wind anomalies over the northern tropical Atlantic from boreal fall and winter, cause the NTA SST anomalies, which grow and peak in spring through a wind, evaporation, and SST (WES) feedback. Most of the initial wind anomalies are in turn generated by the remote ENSO forcing. As a result, the NTA pattern can be predicted skillfully at long leads as long as the ENSO events are predictable. On the other hand, the source of the STA predictability is from the Bjerknes feedback among equatorial zonal wind, SST and thermocline anomalies. Unlike NTA, skillful predictions of the STA events in boreal summer are mostly restricted to initial states during late spring. This is possibly because the strength of the Bjerknes feedback is only strong enough during the transitional season to generate sustainable growth of the initial oceanic anomalies. The ENSO teleconnection does not play a significant role in determining the seasonal predictability of the STA events.